برداشت تفسیر داده های کشف شده در الگو های تکراری جنایات با کاوش fine grained Adopting data interpretation on mining fine-grained near-repeat patterns in crimes

1. Introduction and related works Criminological studies have demonstrated that repeat crimes are essential fundamental phenomenons.1 And the near-repeat effect is widely known because it reveals the elevated tendency between crime incidents taking place nearby in both space and time.2 The major nearrepeat researches concentrate on two aspects. One aspect pays attention to the crimes in particular type.3 The near-repeat phenomenon is first discovered in burglary,4 which is still a hot topic even today.5,6 There are also researches on other single crime types, including robbery,7 shooting8,9 and assault10. 11 assesses the near-repeat phenomenon by Mean Frequencies in three crime types of shooting, robbery and auto theft. But the depth of this research inclines to the unique spatiotemporal pattern of each type, without penetrating deep into the relationships of the types. The other aspect takes notice of the individuals or the relationships among criminal suspects with the help of social network thoughts and research methods.12 utilises epidemiological methods to investigate the phenomenon in the offence of burglary.13 talks about the same offenders involved in near-repeat burglaries.14 analyzes the initiator and near-repeat events in burglary and motor vehicle theft. When it comes to near-repeat phenomenons, we have to mention the importance of locating crime scene in criminal cases, which is the core of the criminal investigations.15 Through the analysis of forensic experts, the police can get the autopsy report,16 the botanical analysis17 and analysis of evidences, such as hanging marks18 and mobile devices.19 In order to make best uses of the information that has been mastered in similar cases, the studies of near-repeat effect and crime patterns are of great significance. From the perspective of crime patterns mining, patterns are based on time or space.20 finds the spatial behavioural patterns of the individual burglar.21 makes the offenses cluster in time, and finds that crimes often occur at particular time. However, given the problems of non temporal or spatial factors, there is no proper method to deal with it. The techniques are usually on the basis of some mathematical distributions, such as, Poisson distribution22 and non-hierarchical clustering method, for example, the k-means clustering.23 However, crime data in the real world rarely matches specific data distribution. And the k-means clustering completely depends on the coordinates of each point, which is limited by Euclidean distance and the mean value. In our paper, we choose hierarchical clustering as a basic step, which is an efficient technique for data mining.